def selection(gs: list, k: int):
gs1 = []
s = 0
for i in gs:
fitness = fitness_function(i)
gs1.append([i, fitness])
s += fitness
gs1.sort(key=lambda x: -x[1])
for i in gs1:
i[1] = (i[1] * 100) / s
g = []
#since, k= 8
for i in range(k // 2):
x = np.random.randint(0, 50)
#selecting first one in pair
for j in gs1:
if x - j[1] < 0:
g.append(j[0])
break
x -= j[1]
#selecting second one in pair
r = g[-1]
while g[-1] == r:
x = np.random.randint(0, 50)
for j in gs1:
if x - j[1] < 0:
r = j[0]
break
x -= j[1]
g.append(r)
return g
def reduction(self): Fit = fitness_function() for x in xrange(0,len(self.descendants)): if self.descendants[x]['value'] > Fit.min and self.descendants[x]['value'] <= Fit.max: self.population.append(self.descendants[x]) self.population.sort(reverse=True) self.population = self.population[:self.pop_size]
def reduction(self):
Fit = fitness_function()
for x in xrange(0, len(self.descendants)):
if self.descendants[x]['value'] > Fit.min and self.descendants[x][
'value'] <= Fit.max:
self.population.append(self.descendants[x])
self.population.sort(reverse=True)
self.population = self.population[:self.pop_size]
def pop_max(self): Fit = fitness_function() print '####################################################' print 'Макс. значение популяции в точке x=', print self.population[0]['value'] print 'f(x)=', print self.population[0]['fitness'],' / ',Fit.optium print '[ min / max ] [', print self.population[-1]['value'],';',self.population[0]['value'],']' print '####################################################' print print
def pop_max(self):
Fit = fitness_function()
print '####################################################'
print 'Макс. значение популяции в точке x=',
print self.population[0]['value']
print 'f(x)=',
print self.population[0]['fitness'], ' / ', Fit.optium
print '[ min / max ] [',
print self.population[-1]['value'], ';', self.population[0][
'value'], ']'
print '####################################################'
print
print
def mutation(self): Fit = fitness_function() for x in xrange(0,len(self.descendants)): chromosome = float_to_binary( self.descendants[x]['value']) gen = random.randint(0,63) if chromosome[gen] == '0': chromosome = chromosome[:gen] + '1' + chromosome[gen+1:] else: chromosome = chromosome[:gen] + '0' + chromosome[gen+1:] self.descendants[x]['value'] = binary_to_float(chromosome) self.descendants[x]['fitness'] = Fit.f(self.descendants[x]['value'])
def genetic_algorithm():
gs = []
k = 30
for i in range(k):
gs.append(random_generation())
for _ in range(1000):
gs1 = selection(gs, k)
for i in range(0, len(gs1), 2):
gs1[i], gs1[i + 1] = crossover(gs1[i], gs1[i + 1])
for i in range(len(gs1)):
gs1[i] = mutation(gs1[i])
gs = gs1[:]
gs.sort(key=lambda x: -fitness_function(x))
m = fitness_function(gs[0])
s = gs[0]
if m == 243:
print(
"5 Best solution along with their fitnesses of the final population"
)
print("Sudoku 1: Solution Found!!")
printSudoku(s)
print("Fitness Value: {}".format(fitness_function(gs[0])))
for j in range(1, 5):
print("Sudoku {}:".format(j + 1))
printSudoku(gs[i])
print("Fitness Value: {}".format(fitness_function(gs[i])))
break
else:
print(
"5 Best solution along with their fitnesses of the final population"
)
for i in range(5):
print("Sudoku {}:".format(i + 1))
printSudoku(gs[i])
print("Fitness Value: {}".format(fitness_function(gs[i])))
print('')
def mutation(self):
Fit = fitness_function()
for x in xrange(0, len(self.descendants)):
chromosome = float_to_binary(self.descendants[x]['value'])
gen = random.randint(0, 63)
if chromosome[gen] == '0':
chromosome = chromosome[:gen] + '1' + chromosome[gen + 1:]
else:
chromosome = chromosome[:gen] + '0' + chromosome[gen + 1:]
self.descendants[x]['value'] = binary_to_float(chromosome)
self.descendants[x]['fitness'] = Fit.f(
self.descendants[x]['value'])
def __init__(self, size, parents_n):
self.pop_size = size
if parents_n > 100:
parents_n = 100
self.parents_num = parents_n
while size > 0:
Fit = fitness_function()
person = {}
person['value'] = Fit.min + random.random() * (Fit.max - Fit.min)
person['fitness'] = Fit.f(person['value'])
size -= 1
self.population.append(person)
self.population.sort(reverse=True)
def __init__(self,size,parents_n):
self.pop_size = size
if parents_n > 100:
parents_n = 100
self.parents_num = parents_n
while size > 0:
Fit = fitness_function()
person = {}
person['value'] = Fit.min + random.random() * (Fit.max-Fit.min)
person['fitness'] = Fit.f(person['value'])
size -= 1
self.population.append(person)
self.population.sort(reverse=True)
def vrp_ga(m=0.2, gen=500, N=100, best_N=5, setup=1, plot_folder="plots/"):
"""
Complete function, optimizing the vehicle routing problem using genetic
algorithms and a custom crossover/mutation part.
Args
m mutation probability
gen number of generation (iterations)
N population size (must be divisable by 4)
best_N plotting the N best individuals each generation
setup number of setup (currently 1 and 2)
plot_folder path to save plots
"""
if N % 4 != 0:
raise ValueError(
"[!] 'N' (population size) has to be dividedable by 4!")
history = np.zeros((gen, best_N))
# create one test member, to load task
# cap = capacity of trucks in list
# demands = demands of cities
# dist = distance matrix
# tc = transportation costs of trucks
_, cap, demands, dist, tc = complete_init(task_nr=setup)
##### initialization
# create a list for our population with N elements, with multiple
# possible setups as numpy arrays
# [instead of doing a list of 2d arrays, we do one 3d array, where the
# first dimension stands for the members]
# population has 3 dimensions: members x trucks x cities
population = np.zeros((N, len(cap), len(demands)))
# fill the population array
for i in range(N):
population[i] = complete_init(task_nr=setup)[0]
for gen_idx in tqdm(range(gen)):
##### fitness
permutations = np.array(
[permute_way(member, dist, demands, cap) for member in population])
fitness_scores = np.array([
fitness_function(dist, permutation, member, tc)
for permutation, member in zip(permutations, population)
])
history[gen_idx] = sorted(fitness_scores)[-best_N:]
##### selection
selected_indx = roulette_wheel_selection(fitness_scores, int(N / 2))
selected_members = population[selected_indx]
np.random.shuffle(selected_members)
##### crossover
children = vrp_crossover(selected_members, cap, demands)
##### mutation
for i, child in enumerate(children):
if random.uniform(0, 1) < m:
children[i] = mutate(child, cap)
##### replacement
population = replacement(population,
children,
mode="delete-all",
n=len(children),
based_on_fitness=True,
fitness_old=fitness_scores)
##### plotting
plt.figure()
plt.plot(history)
plt.xlabel("Generations")
plt.savefig(plot_folder + "history" +
strftime("%Y-%m-%d %H:%M:%S", gmtime()) + ".png")
plt.show()
